U.S. patent application number 16/800645 was filed with the patent office on 2020-06-18 for determining a load status of a platform using a likelihood ratio test.
The applicant listed for this patent is BlackBerry Limited. Invention is credited to Dake He, Victor James Stolpman.
Application Number | 20200192972 16/800645 |
Document ID | / |
Family ID | 60484497 |
Filed Date | 2020-06-18 |
![](/patent/app/20200192972/US20200192972A1-20200618-D00000.png)
![](/patent/app/20200192972/US20200192972A1-20200618-D00001.png)
![](/patent/app/20200192972/US20200192972A1-20200618-D00002.png)
![](/patent/app/20200192972/US20200192972A1-20200618-D00003.png)
![](/patent/app/20200192972/US20200192972A1-20200618-D00004.png)
![](/patent/app/20200192972/US20200192972A1-20200618-D00005.png)
![](/patent/app/20200192972/US20200192972A1-20200618-M00001.png)
![](/patent/app/20200192972/US20200192972A1-20200618-M00002.png)
![](/patent/app/20200192972/US20200192972A1-20200618-M00003.png)
![](/patent/app/20200192972/US20200192972A1-20200618-M00004.png)
![](/patent/app/20200192972/US20200192972A1-20200618-M00005.png)
View All Diagrams
United States Patent
Application |
20200192972 |
Kind Code |
A1 |
Stolpman; Victor James ; et
al. |
June 18, 2020 |
DETERMINING A LOAD STATUS OF A PLATFORM USING A LIKELIHOOD RATIO
TEST
Abstract
In some examples, measurement data is received from at least one
sensor that detects a signal reflected from a surface inside a
platform. A likelihood ratio test is applied using the measurement
data, and a load status of the platform is determined based on the
likelihood ratio test.
Inventors: |
Stolpman; Victor James;
(Spring, TX) ; He; Dake; (Waterloo, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
BlackBerry Limited |
Waterloo |
|
CA |
|
|
Family ID: |
60484497 |
Appl. No.: |
16/800645 |
Filed: |
February 25, 2020 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15350488 |
Nov 14, 2016 |
10572570 |
|
|
16800645 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06Q 10/083 20130101;
G01V 8/10 20130101; G01S 17/04 20200101; G06Q 50/28 20130101; G06F
17/18 20130101 |
International
Class: |
G06F 17/18 20060101
G06F017/18; G06Q 10/08 20060101 G06Q010/08; G01S 17/04 20060101
G01S017/04; G01V 8/10 20060101 G01V008/10 |
Claims
1. A method performed by at least one processor, comprising:
receiving measurement data from at least one sensor that detects a
signal reflected from a surface inside a platform; applying a
likelihood ratio test using the measurement data; and determining a
load status of the platform based on the likelihood ratio test.
2. The method of claim 1, wherein applying the likelihood ratio
test comprises computing a value using an estimated baseline
corresponding to an empty platform.
3. The method of claim 2, wherein applying the likelihood ratio
test further comprises: using the measurement data from the at
least one sensor as a representation of a loaded platform;
comparing the computed value with respect to at least one
threshold, wherein determining the load status of the platform is
based on the comparing.
4. The method of claim 3, wherein comparing the computed value with
respect to the at least one threshold comprises comparing the
computed value with respect to a first threshold and a second
threshold, and wherein the determining comprises: indicating that
the platform is empty responsive to the computed value being less
than the first threshold, and indicating that the platform is
loaded responsive to the computed value exceeding the second
threshold.
5. The method of claim 1, wherein applying the likelihood ratio
test comprises applying a log likelihood ratio test.
6. The method of claim 1, wherein applying the likelihood ratio
test comprises: computing a plurality of values using a likelihood
ratio test function applied on corresponding samples of measurement
data received from the at least one sensor; aggregating the
plurality of values to produce an aggregate value, wherein
determining the load status of the platform is based on the
aggregate value.
7. The method of claim 6, wherein determining the load status of
the platform is based on comparing the aggregate value to at least
one threshold, wherein the aggregate value having a first
relationship to the at least one threshold indicates that the
platform is empty, and the aggregate value having a second
relationship to the at least one threshold indicates that the
platform is loaded.
8. The method of claim 7, further comprising: computing the at
least one threshold based on tuning parameters selected to control
error rates.
9. The method of claim 1, wherein applying the likelihood ratio
test comprises applying a generalized likelihood ratio test.
10. The method of claim 9, wherein applying the generalized
likelihood ratio test comprises: defining a subspace based on a
phase; identifying a point in the subspace that is closest to
points representing measurement data received from the at least one
sensor; using the identified point as an estimated baseline
corresponding to an empty platform; computing a value using the
estimated baseline.
11. The method of claim 10, wherein computing the value using the
estimated baseline comprises computing the value using a
log-likelihood ratio test or a sequential probability ratio
test.
12. An apparatus comprising: at least one sensor to output
measurement data responsive to detecting a signal reflected from a
surface inside a platform; at least one processor configured to:
compute a value by applying a likelihood ratio test function on the
measurement data; and determine a load status of the platform based
on the computed value.
13. The apparatus of claim 12, wherein the at least one sensor
comprises a light detector to detect reflected light emitted by a
light emitter.
14. The apparatus of claim 12, wherein the likelihood ratio test
function is based on a ratio between a model representing an empty
platform and a model representing a loaded platform.
15. The apparatus of claim 14, wherein the model representing the
empty platform is based on a mean of measurement data samples
representing the empty platform, and the model representing the
loaded platform is based on a mean of measurement data samples
representing the loaded platform.
16. The apparatus of claim 12, wherein the at least one processor
is configured to: compute a plurality of values output by applying
the likelihood ratio test function on a plurality of measurement
data samples from the at least one sensor; and aggregate the
plurality of values to produce an aggregate value, wherein the
determining of the load status is based on comparing the aggregate
value to at least one threshold.
17. The apparatus of claim 12, wherein the at least one processor
is configured to compare the computed value to at least one
threshold, wherein the computed value having a first relationship
to the at least one threshold indicates that the platform is empty,
and the computed value having a second relationship to the at least
one threshold indicates that the platform is loaded.
18. The apparatus of claim 12, wherein the at least one processor
is configured to: determine an estimated baseline corresponding to
an empty platform by: identifying a point in a subspace that is
closest to points representing measurement data received from the
at least one sensor, wherein the subspace corresponds to a
specified phase in In-phase/Quadrature (IQ) space, wherein the
likelihood ratio test function is based on the estimated
baseline.
19. A non-transitory machine-readable storage medium storing
instructions that upon execution cause a system to: receive
measurement data from at least one sensor that detects a signal
reflected from a surface inside a platform; apply a likelihood
ratio test using the measurement data, the likelihood ratio test
producing an output value; and determine a load status of the
platform based on comparing the output value to at least one
threshold.
20. The non-transitory machine-readable storage medium 19, wherein
the determining comprises: indicating that the platform is empty
responsive to the output value having a first relationship with
respect to the at least one threshold, and indicating that the
platform is loaded responsive to the output value having a second
relationship with respect to the at least one threshold.
Description
CROSS REFERENCE TO RELATED APPLICATIONS
[0001] This is a continuation of U.S. application Ser. No.
15/350,488, filed Nov. 14, 2016, which is hereby incorporated by
reference.
BACKGROUND
[0002] Trucks, tractor-trailers, or tractors that are connected to
chassis for carrying containers can be used to transport cargo that
includes goods. Trucks, tractor-trailers, and containers typically
have doors that can be opened to allow access to cargo that is
being transported, and closed to secure the cargo.
BRIEF DESCRIPTION OF THE DRAWINGS
[0003] Some implementations of the present disclosure are described
with respect to the following figures.
[0004] FIGS. 1A-1B are schematic diagrams of a container hauled by
a vehicle, the container including a door and a sensor device
according to some implementations.
[0005] FIG. 2 is a block diagram of an example sensor device
according to some implementations.
[0006] FIG. 3 is a flow diagram of an example process according to
some implementations.
[0007] FIG. 4 a block diagram of an example time-of-flight sensor
according to some implementations.
[0008] FIGS. 5A and 5B are graphs plotting output values computed
by likelihood ratio tests as a function of measurement samples,
according to some implementations.
[0009] FIG. 6 is a graph of an in-phase/quadrature-phase plane
showing a subspace used for a likelihood ratio test, according to
further implementations.
DETAILED DESCRIPTION
[0010] In the present disclosure, use of the term "a," "an", or
"the" is intended to include the plural forms as well, unless the
context clearly indicates otherwise. Also, the term "includes,"
"including," "comprises," "comprising," "have," or "having" when
used in this disclosure specifies the presence of the stated
elements, but do not preclude the presence or addition of other
elements.
[0011] A moveable platform can be used to carry physical items
between different geographic locations. For example, the moveable
platform can be a container (that is attached to a tractor), a
truck, or a trailer, where the container provides an enclosed space
in which the physical items can be stored during shipment. In other
examples, the moveable platform can include another type of carrier
structure that is able to carry physical items. More generally, the
moveable platform can be part of, mounted on, or attached, as
applicable, to a vehicle, such as a truck, a trailer, a tractor, a
car, a train, a ship, an airplane, and so forth. In the ensuing
discussion, a moveable platform can be referred to as a container.
It is noted that techniques or mechanisms according to some
implementations can be applied to other cargo carrying platforms
with an entry barrier that can be opened and closed.
[0012] A moveable platform can include a door through which
physical items can be loaded or unloaded into or from an inner
chamber of the moveable platform, respectively. The door is an
example of an entry barrier (or more simply, a "barrier") that can
be opened and closed. Other examples of barriers include a window
or any other structure that can be opened to allow entry through an
opening, or closed to block entry through the opening
[0013] An entity such as a shipper, a distributor, a manufacturer,
a seller of goods, or any other entity may wish to track assets
(such as cargo) that are being transported using moveable
platforms. To do so, a sensor device can be mounted on a moveable
platform. Sensor devices on various moveable platforms can
communicate sensor information over a network to a remote service
to allow the remote service to track assets that are being moved by
various moveable platforms. The remote service can include a server
or a collection of servers and associated network equipment that
may be located at one fixed location or in a mobile unit or as part
of a data center or cloud. Asset tracking can involve tracking the
current locations of the assets, cargo load status of moveable
platforms, conditions of the environment around the assets (where
such conditions can include a measured temperature, a measured
humidity, etc.), and/or other information.
[0014] A sensor device can include a communication component to
communicate over a network. In some examples, sensor devices
mounted on moveable platforms can be part of a larger network of
devices. This larger network of devices can be part of the
"Internet-of-Things" (IoT) technology paradigm to allow different
types of devices to communicate different types of data (including
sensor data, voice data, video data, e-mail data, picture or image
data, messaging data, web browsing data, and so forth). In addition
to network traffic communicated by computers, smartphones, wearable
devices, and the like, the IoT technology paradigm contemplates
that other types of devices, including household appliances,
vehicles, sensor devices, thermostats, and so forth have
connectivity to a network to allow the devices to communicate
respective data.
[0015] More generally, asset tracking using network connected
sensor devices can involve acquiring sensor data, transmitting the
sensor data, and aggregating such sensor data or producing other
measures based on the sensor data to determine information
associated with the assets that are being transported by moveable
platforms. Based on data received from the sensor devices, a server
(or servers) can update a database, run analytics, and/or present
status information for display, such that further decisions or
actions can be performed. The asset tracking can be used to improve
fleet utilization, reduce operating cost, reduce loss of assets due
to theft, and so forth.
[0016] In some examples, a measure that can be derived based on
measurements made by a sensor device mounted on a moveable platform
is a load status of the moveable platform, where the load status
can refer to whether or not the moveable platform contains any
physical items such as cargo (i.e., the container is empty or the
container contains at least one physical item). More generally,
determining a load status can refer to an amount of loading (e.g.,
0% loaded, 10% loaded, 25% loaded, 50% loaded, 100% loaded, or by
use of some other measure that can have different values that
indicate different levels of loading with physical items.
Accurately detecting the load status of a container can be
challenging due to one or more factors. For example, the
characteristics of a moveable platform may change over time, such
as due to addition of structures within the chamber of the moveable
platform, presence of debris in the moveable platform chamber,
cleaning of the inner chamber of the moveable platform, repainting
of the inner walls of the moveable platform, and so forth. In
addition, the hardware of the sensor device can also vary over
time, such as due to varying temperature or due to aging of the
hardware, which can affect the sensitivity, gain, and/or other
characteristic of a sensor in the sensor device. Moreover,
different moveable platforms and different sensor devices may have
different characteristics due to variances in manufacturing and/or
configuration. Furthermore, the size and construction composition
of the moveable platform can vary from one moveable platform to
another. Also, the surrounding environment for different sensor
devices may be different. Due to such varying characteristics of
the moveable platform and/or the sensor device, a determination of
a load status based on measurement data from the sensor device may
not be accurate.
[0017] Additionally, detection systems for detecting a load status
of a moveable platform that rely upon measurements of physical
distances within the moveable platform to determine if cargo is
present may produce sub-optimal results. Sub-optimal detection
systems may miss the physical items within a moveable platform, or
may provide a false indication that the moveable platform is loaded
with a physical item.
[0018] In accordance with some implementations of the present
disclosure, statistical techniques are applied to determine a load
status of a moveable platform. In some implementations, the
statistical techniques include likelihood ratio test techniques,
where a likelihood ratio test generally refers to a statistical
test that is used to compare the goodness of fit of two models,
including a first model that is a null model and a second model
that is an alternative model. The test is based on a likelihood
ratio between the null model representing an empty container and
the alternative model representing a loaded container, which
expresses how much more likely data fits the null model versus the
alternative model. In the context of detecting the load status of a
moveable platform, the null model (or null hypothesis) represents a
moveable platform is empty, while the alternative model, or
alternative hypothesis, represents the moveable platform including
at least one physical item (i.e., the moveable platform is
loaded).
[0019] In some examples, a likelihood ratio test can include a log
likelihood ratio (LLR) test. Another likelihood ratio test is a
sequential probability ratio test (SPRT), which employs the LLR
test to produce multiple LLR outputs for respective multiple
measurement data samples. The multiple measurement data samples can
include a sequence of measurement data samples collected from at
least one sensor for detecting physical items in a moveable
platform. The SPRT technique can aggregate (e.g., sum) the multiple
LLR outputs values to produce an SPRT value that can be used for
determining the load status of a moveable platform. In further
examples, another likelihood ratio test technique that can be
applied is a general likelihood ratio test (GLRT), which can be
applied in instances where certain information relating to a
moveable platform is missing, such as the color of the moveable
platform or other information relating to a characteristic of the
moveable platform. The GLRT technique can thus be applied in cases
where partial information regarding characteristics of the moveable
platform are available.
[0020] FIGS. 1A-1B illustrate an example truck 100 that includes a
tractor unit 102 and a container 104 (provided on a chassis) hauled
by the tractor unit 102. The container 104 is an example of a
moveable platform that can be used to carry physical items. FIG. 1B
is a perspective view of the container 104, to show that the
container 104 has a door 106 (or other entry barrier) that is
pivotable between an open position and a closed position (or more
generally, moveable between an open position and a closed
position).
[0021] In FIG. 1B, the door 106 is in the open position. The door
106 is pivotably mounted to hinges 108, which are attached to a
door frame 110 of the container 104. The door 106 is able to rotate
about the hinges 108 between the open position and the closed
position. Although FIG. 1B shows two hinges 108, it is noted that
in other examples, the door 106 can be mounted on just one hinge,
or more than two hinges.
[0022] In some examples, a sensor device 112 is mounted to an
interior surface of the door 106, such that when the door 106 is
closed, the sensor 112 faces towards an inner chamber 114 of the
container 104. As noted above, the container 104 is an example of a
moveable platform. The inner chamber 114 of the container 104
(referred to as a "container chamber" in the ensuing discussion) is
an example of an inner space of the moveable platform in which
physical items can be carried. The container chamber 114 is defined
by the walls of the container 104. In other examples, the sensor
device 112 is not mounted on the door 106, but rather can be
mounted on another surface inside the container 104.
[0023] In some implementations, in response to a door-closing
event, the sensor device 112 can make a measurement for use in
determining the load status of the container 104, e.g., to
determine whether there is any physical item located within the
container chamber 114 of the container 104, or to determine a level
of loading of physical items within the container chamber 114. For
example, the sensor device 112 includes a Time-of-Flight (ToF)
sensor 116.
[0024] In some examples, the determination of the load status of
the container 104 can be performed by a processor (or multiple
processors) in the sensor device 112. In other examples, the
determination of the load status of the container 104 can be
performed by a remote server computer (or multiple remote server
computers) that receive the measurement data from the sensor device
112.
[0025] In FIG. 1A, it is assumed that physical items 118 are
located within the container chamber 114. The ToF sensor 116 is
able to emit a signal 113 that is reflected from a surface inside
the container chamber 114. Based on the detection of the reflected
signal by the ToF sensor 116, the sensor device 112 outputs
measurement data that can be used in likelihood ratio test
techniques to determine the load status of the container 104.
[0026] In examples where the ToF sensor 116 includes a light
emitter, such as a light emitting diode (LED), a transmitted
periodic signal can include multiple light pulses that employ an
on-off-keying (OOK) of the light emitter. In other examples, other
types of transmitted periodic signals can be employed. The ToF
sensor 116 includes a light detector to detect light reflected by
the light emitter.
[0027] In some examples, a light source such as the light emitter
mentioned above may be modulated by a modulation signal. The
frequency of the modulation signal can be used in converting the
phase shift of the reflected light to distance measure. In some
examples, the modulation frequency can be in a range between 2
megahertz (MHz) to or 5 MHz to more than 10 MHz, depending upon the
ranges of distance to be detected.
[0028] In the FIG. 1A example, the emitted signal 113 is reflected
from a rear-facing surface 120 of the rear-most physical item 118.
The rear-facing surface 120 faces the door 106 of the container
104. In examples where there are multiple physical items, the ToF
sensor 116 detects a reflected signal from the physical item 118
that is closest to the door 106.
[0029] If the physical items 118 are not present in the container
chamber 114, then the ToF sensor 116 detects a signal reflected
from an inner front surface 122 of the container 104.
[0030] The inner front surface 122 of the container 104 can be the
front-most wall of the container 104. In other examples, another
structure such as a frame can be located in the container chamber
114 against which physical items 118 can be placed. In such other
examples, the inner front surface 122 is a surface of such other
structure, or the floor of the container 104.
[0031] Although the present discussion refers to examples where the
door 106 is located at the rear of the container 104 and physical
items 118 are loaded from the front of the container chamber 114 to
the rear of the container chamber 114, it is noted that in other
examples, the door 106 can be located at a different part of the
container 104, such as on a side wall of the container 104, or on a
front wall of the container 104. In such other examples, the
loading of physical items can be from side to side, or from the
rear to the front. The sensor device 112 can also be located on a
side wall or the front wall of the container 104 in such other
examples.
[0032] In the foregoing discussion, it is assumed that measurement
data from the sensor device 112 is used to determine the load
status of the entirety of the container chamber 114. In other
examples, the container chamber 114 can be divided into multiple
zones, and the sensor device 112 (or multiple sensor devices 112)
can be used to determine the load status of each zone of these
multiple zones. Thus, the sensor device(s) 112 can determine a load
status of a first zone, a load status of a second zone, and so
forth.
[0033] FIG. 2 is a block diagram of the sensor device 112 according
to some examples. The sensor device 112 includes the ToF sensor
116, a processor (or multiple processors) 202, and a communication
component 204. A processor can include a hardware processing
circuit, such as a microprocessor, a core of a multi-core
microprocessor, a microcontroller, a programmable gate array, a
programmable integrated circuit device, or any other hardware
processing circuit.
[0034] The sensor device 112 further includes a battery 210 that
provides power to components of the sensor device 112.
[0035] In some examples, the processor 202 receives measurement
data from the ToF sensor 116, and based on the measurement data,
the processor 202 is able to determine a load status of the
container 104 (either the load status of the entirety of the
container, or the load status of a zone of multiple zones in the
container). The processor 202 provides the load status indication
to the communication component 204, which transmits the load status
indication 206 over a network to a destination, such as a remote
service that is used to perform asset tracking.
[0036] The communication component 204 can include a wireless
transceiver and associated circuits to allow for wireless
communication of output data 206 by the sensor device 112 to the
network. The wireless communication can include wireless
communication over a cellular access network, a wireless local area
network, a satellite network, and so forth. Alternatively, the
communication component 204 can include a wired transceiver and
associated circuits to perform wired communications between the
sensor device 108 and the destination.
[0037] In other examples, the sensor device 112 does not perform
the load status determination, but rather, the sensor device 112
transmits measurement data based on ToF data of the ToF sensor 116
using the communication component 204 to the destination. In such
examples, the determination of the load status can be performed by
a server (or servers) at the destination, based on the measurement
data transmitted by the sensor device 112. In such examples, the
processor 202 can simply transfer measurement data received from
the ToF sensor 116 to the communication component 204 for
transmission to the destination.
[0038] In examples where the processor 202 of the sensor device 112
is configured to perform the load status determination, the
processor 202 can execute a load status detector 208, which
includes software or other machine-readable instructions executable
on the processor 202. In other examples, the load status detector
208 can be implemented using hardware. The load status detector 208
determines, based on measurement data from the ToF sensor 116, the
load status of the container 104.
[0039] In some examples, the load status detector 208 includes a
likelihood ratio test (LRT) function 209 that is used to apply a
likelihood ratio test using measurement data from the ToF sensor
116 for determining the load status of the container 104. The LRT
function 209 can be invoked by the load status detector 208 to
apply any of various likelihood ratio test techniques according to
some examples, such as the LLR test technique, SPRT technique, or
GLRT technique, which are described in greater detail below.
[0040] Although the LRT function 209 is shown as being part of the
load status detector 208 in FIG. 2, it is noted that in other
examples, the LRT function 209 can be separate from the load status
detector 208.
[0041] In examples where the sensor device 112 does not perform the
load status determination, the load status detector 208 and LRT
function 209 can be provided at a server (or multiple servers) that
is (are) remote from the sensor device 112. Shifting the
determination of the load status of a container to a remote
server(s), instead of performing the load status determination at
the sensor device 112, can reduce the complexity of the sensor
device 112, and can allow the sensor device 112 to consume less
power to save battery power in the sensor device 112.
[0042] In some examples, to reduce power consumption of the sensor
device 112, and thus to conserve the battery power, the sensor
device 112 can be maintained in a sleep state until an event
triggers the sensor device 112 to make a measurement and/or to
perform processing tasks. A sleep state refers to a state of the
sensor device 112 where the sensor device is powered off, or a
portion of the sensor device 112 is powered off, such that the
sensor device 112 consumes a lower amount of power than another
state of the sensor device, such as an operational state. An
operational state of the sensor device 112 is a state of the sensor
device 112 where the sensor device is able to perform specified
tasks, including measurement of data and/or processing of data. In
the operational state, the sensor device 112 consumes more power
than the power consumed by the sensor device in the sleep
state.
[0043] In some examples, an event that can trigger the sensor
device 112 to transition from the sleep state to the operational
state can be a door close event, which is generated when the door
106 is closed from an open position. The sensor device 112 can
include a door status sensor 212 to detect either the opening or
closing of the door 106 (FIG. 1) or other entry barrier. Although
FIG. 2 shows the door status sensor 212 as being part of the sensor
device 112, it is noted that in other examples, the door status
sensor 212 can be external of the sensor device 112.
[0044] The door status sensor 212 can detect a change in status of
the door 106 using any of various mechanisms. For example, a switch
can be attached to the door, where the switch changes state in
response to the door being opened or closed. As another example, a
magnetic sensor can be used, where the magnetic sensor can be in
proximity to a magnet when the door is closed, but when the door is
opened, the magnetic sensor moves away from the magnet. The
magnetic sensor can thus output different values depending upon
whether the door is opened or closed. In other examples,
acceleration data from an accelerometer and rotation data from a
rotation sensor (such as a gyroscope or rotation vector sensor) can
be used for detecting the door being opened and closed.
[0045] In response to the door status sensor 212 indicating that
the door has been closed from an open position, the sensor device
112 can be awakened from the sleep state to the operational state.
For example, the ToF sensor 116 can be activated from a lower power
state to a higher power state, and/or the processor 202 can be
activated from a lower power state to a higher power state. In some
examples, in response to detecting that the door has been closed,
the sensor device 112 can wait a specified time duration before
transitioning from the sleep state to the operational state, such
as to avoid triggering multiple transitions between the sleep state
and the operational state in a short period of time, such as due to
a person opening and closing the door in quick succession.
Additionally, the processor 202 can wait to transition from the
lower power state to the higher power state to allow the ToF sensor
116 time to power on and initialize and take measurements.
[0046] FIG. 3 is a flow diagram of an example process for deciding
whether a moveable platform such as the container 104 contains a
physical item (or physical items) such as cargo. The process of
FIG. 3 can be performed by the processor 202 in the sensor device
112, or alternatively, by a remote server.
[0047] The process of FIG. 3 includes receiving (at 302)
measurement data from at least one sensor (e.g., the ToF sensor
116) that detects a signal reflected from a surface inside the
moveable platform. The process of FIG. 3 further includes applying
(at 304) a likelihood ratio test using the measurement data. The
process determines (at 306) a load status of the moveable platform
based on the likelihood ratio test.
[0048] The likelihood ratio test that is applied using the
measurement data can produce a computed value, which can be
compared to one or more thresholds for determining the load status
of the moveable platform. For example, if the computed value has a
first relationship with respect to the one or more thresholds, then
the moveable platform is indicated as being empty. On the other
hand, if the computed value has a second relationship with respect
to the one or more thresholds, then the moveable platform is
indicated as being loaded. In some examples, the first relationship
can be a "less than" or "less than or equal" relationship, while
the second relationship can be a "greater than" or "greater than or
equal" relationship. In other examples, other relationships can be
employed.
[0049] In some examples, the measurement data from the at least one
sensor can be used as a representation of a loaded platform, and
the computed value from the likelihood ratio test can be compared
with respect to at least one threshold. Determining the load status
of the platform is based on the comparing.
[0050] In some examples, multiple thresholds can be used, including
a lower threshold and an upper threshold. The computed value
calculated based on applying the likelihood ratio test on
measurement data if less than the lower threshold provides an
indication that the moveable platform is empty, while the computed
value being greater than the upper threshold provides an indication
that the moveable platform is loaded with at least one physical
item. In other examples, just one threshold can be used.
[0051] FIG. 4 is a block diagram of an example ToF sensor 116,
which includes a light emitter 402 and a light detector 404.
Although just one light emitter 402 and one light detector 404 are
shown in FIG. 4, the ToF sensor 116 in other examples can include
multiple light emitters and/or multiple light detectors. The light
emitter 402 can include a light emitting diode (LED) or another
type of light source that can produce light either in the visible
spectrum or invisible spectrum (e.g., infrared or ultraviolet
light). The light detector 404 can include a photo-sensitive diode
or other type of light detector.
[0052] The light emitter 402 can be used to transmit an emitted
light signal 406, which can include one or more light pulses in
some examples, into the container chamber 114. An input voltage 401
is applied to the light emitter 402 to cause the light emitter 402
to transmit the emitted light signal 406. The emitted light signal
406 is reflected as a reflected light signal 408 from a surface
inside the container chamber 104, where the surface can be the
rear-facing surface 120 of a physical item 118, or the front
surface 122 of the container 104 (FIG. 1). The reflected light
signal 408 is detected by the light detector 404, which produces an
output 410 that is responsive to the reflected light signal 408.
The output 410 can include an output voltage or an output current,
which has a value (e.g. an amplitude) that is based on a property
of the reflected light signal 408. Although not shown, the output
410 can be provided through a signal chain including intermediate
circuits, such as an analog-to-digital (ADC) converter, an
amplifier, a buffer, and so forth.
[0053] In other examples, the ToF sensor 116 can include a
different type of signal emitter to emit another type of signal,
such as an acoustic signal, electromagnetic signal, and so forth.
In such other examples, the ToF sensor 116 can also include a
different type of signal detector, such as an acoustic sensor, an
electromagnetic sensor, and so forth.
[0054] The following provides further details associated with
various likelihood ratio tests can be employed for determining a
load status of a moveable platform.
[0055] In the ensuing discussion, a signal transmitted by the ToF
sensor 116, such as by the light emitter 402 or other signal
emitter, can be represented as x(t). A signal received by the ToF
sensor 116, such as by the light detector 404 or other signal
sensor, can be represented as y(t).
[0056] In more specific examples, x(t) can be a transmitted signal
of N pulse trains including M pulses per pulse train. Note that in
other examples, the transmitted signal x(t) can have a form
different from that described below.
[0057] For each pulse train of M pulses, T separates each of the
individual pulses of width T/2. Thus a single pulse train of M
pulses resembles a square-wave signal with period T at 50% duty
cycle for 0.ltoreq.t<MT (M multiplied by T) and zero for all
other times t. By spacing each pulse train T.sub.s seconds apart
(T.sub.s represents the length of a sampling period), where
MT<T.sub.s, a continuous-time transmitted signal, x(t), can be
expressed as:
x ( t ) = .DELTA. k = 0 N - 1 P p T / 2 ( t ) * .delta. T M ( t -
kT s ) , ( Eq . 1 ) ##EQU00001##
where {square root over (P)} represents a pulse amplitude, and
other parameters of Eq. 1 are provided below. The operator *
denotes continuous-time linear convolution defined as
u(t)*v(t).intg..sub.-.infin..sup.+.infin.u(s)v(t-s) ds. (Eq.2)
[0058] Let .delta.(t) be a Dirac delta function, i.e.,
.delta. ( t ) = .DELTA. lim a -> 0 1 a .pi. e - t 2 / a 2 . ( Eq
. 3 ) ##EQU00002##
[0059] Let .delta..sub.T.sup.M(t) be a waveform including a train
of M Dirac delta functions time-shifted and spaced T seconds
apart:
.delta. T M ( t ) = .DELTA. m = 0 M - 1 .delta. ( t - mT ) . ( Eq .
4 ) ##EQU00003##
[0060] Let p.sub.w(t) be a single rectangular pulse of amplitude 1,
width W, starting at t=0, ending at t=W, and amplitude zero
otherwise, i.e.,
p W ( t ) = .DELTA. { 1 ; 0 .ltoreq. t < W 0 ; otherwise . ( Eq
. 5 ) ##EQU00004##
[0061] As noted above, the likelihood ratio test compares the
goodness or fit of two models, a null model (or null hypothesis)
and an alternative model (or alternative hypothesis).
[0062] Let H.sub.0 be the null hypothesis representing an empty
container, i.e., no physical item. For H.sub.0, the received signal
y.sub.0(t) is expressed as:
y.sub.1(t)=.varies..sub.cA.sub.cx(t-T.sub.c)+.eta.(t), (Eq. 6)
where .varies..sub.c is the reflection coefficient of the
container, A.sub.cis the effective reflective area of the sensor
device's field of view cone for an empty container, T.sub.c is the
round-trip time shift attributed to the container's reflection, and
.eta.(t) is additive noise.
[0063] Let H.sub.1 be the alternative hypothesis representing at
least one physical item being present within the container. For the
alternative hypothesis, the received signal y.sub.1(t) is expressed
as:
y.sub.1(t)=.varies..sub.bA.sub.bx(t-T.sub.b)+.varies..sub.c(A.sub.c-A.su-
b.b)x(t-T.sub.c)+.eta.(t), (Eq. 7)
where .varies..sub.b be the reflection coefficient of the physical
item, A.sub.b is the effective reflective area of the physical
item, and T.sub.b is the round-trip time shift attributed by the
physical item's reflection.
[0064] In standard form, the hypothesis test is
H.sub.0: y(t)=y.sub.0(t) (Empty) (Eq. 8)
H.sub.1: y(t)=y.sub.1(t) (Not Empty),
where y(t) is the received signal corresponding to the transmitted
x(t), y.sub.0(t) is the received signal corresponding to an empty
container, and y.sub.1(t) is the received signal corresponding to a
container that is not empty, i.e., a physical item is present. As
written, the decision is binary where 0 means empty and 1 means not
empty (i.e., no rejection region as written). In other examples,
the hypothesis test can be generalized to indicate more than two
different levels of loading of the container.
[0065] In some examples, before applying the hypothesis test noted
above, a receiver can first multiply a received signal y(t) by
in-phase and quadrature-phase (I/Q) square-waveforms, and integrate
before sampling using an analog-to-digital converter (ADC).
y.sub.1(t)=y(t){tilde over (x)}(t) (In-Phase), (Eq. 9)
y.sub.Q(t)=y(t){tilde over (x)}(t-T/4) (Quadrature). (Eq. 10)
[0066] Due to a common DC component, the pair x(t) and x(t-T/4) do
not form an orthogonal basis and adds unnecessary mathematical
explanation burden to the phase interpretation. For a more
pseudo-intuitive discussion about phase using coordinates on an I/Q
plot, the DC components and nuisance scalar can be removed as
follows:
x ~ ( t ) = .DELTA. 2 P x ( t ) - 1 = 2 k = 0 N - 1 p T / 2 ( t ) *
.delta. T M ( t - kT s ) - 1 x ~ ( t - T / 4 ) = .DELTA. 2 P x ( t
- T / 4 ) - 1 = 2 k = 0 N - 1 p T / 2 ( t - T / 4 ) * .delta. T M (
t - kT s ) - 1 ( Eq . 11 ) ##EQU00005##
According to Eq. 11, an orthogonal basis can be provided in the
form of zero-mean square-waves emulating the sine and cosine
waveforms. Note also that {tilde over (x)}(t) and {tilde over
(x)}(t-T/4) still have a 50% duty-cycle but now they are antipodal
between 1 and -1 amplitude with zero mean over integer multiples of
T.
[0067] Note that scaling both {tilde over (x)}(t) and {tilde over
(x)}(t -T/4) by 1/ {square root over (MT)} produces an orthonormal
basis, {tilde over (x)}(t)/ {square root over (MT)} and {tilde over
(x)}(t-T/4 {square root over (MT)},
[0068] In other examples, an orthogonal basis exists for when the
transmitted waveform resembles a sinusoidal waveform. This may be
intentional and may be done by filtering x(t) with a low-pass
filter with a cutoff frequency above 1/T and below an integer
number of harmonics, or it may be unintentional due to the
bandwidth restrictions of hardware behaving as a low-pass filter.
In these examples, x(t) and x(t -T/4) are sinusoidal waveforms
90.degree. out of phase (e.g., cosine and sine waveforms with a
frequency of 1/T hertz) and form a basis corresponding to the I/Q
coefficients.
[0069] Next, the receiver then integrates the In-phase &
Quadrature (I/Q) signals over NT for each of the N pulse trains
comprising of M pulses, i.e., the discrete I/Q signals after
integration and sampling can be written as
y.sub.1[n]=.intg..sub.nT.sub.s.sup.nT.sup.s.sup.+MTy(t){tilde over
(x)}(t) dt, for n=0,1, . . . , (N+1) (In-Phase) (Eq. 11)
y.sub.Q[n]=.intg..sub.nT.sub.s.sup.nT.sup.s.sup.+MTy(t){tilde over
(x)}(t-T/4) dt, for n=0,1, . . . , (N-1) Quadrature) (Eq. 12)
[0070] With the following assumptions, the null hypothesis and
alternative hypothesis can be expressed according to Eqs. 13-16
further below.
[0071] In some examples, the assumptions are as follows:
[0072] 1) A physical item should be closer than the farthest wall
of the container and is within sensor range, i.e.
0.ltoreq.T.sub.b<T.sub.c<T/2.
[0073] 2) The integration time is less than the sampling period,
i.e., 0<MT<T.sub.s.
[0074] 3) The reflection coefficients are bounded between 0 (black)
and 1(white), i.e., 0.ltoreq..varies..sub.c.ltoreq.1 and
0.ltoreq..varies..sub.b.ltoreq.1.
[0075] 4) The reflective area of the physical item is bounded by
the area of the container, i.e.,
0.ltoreq.A.sub.b.ltoreq.A.sub.c.
[0076] 5) The time shift attributed to the physical item is bounded
by the empty container, i.e., 0.ltoreq.T.sub.b.ltoreq.T.sub.c.
[0077] 6) Noise sampling is performed as follows:
.eta. I [ n ] = .DELTA. .intg. nT s nT S + MT .eta. ( t ) x ~ ( t )
dt , .eta. Q [ n ] = .DELTA. .intg. nT s nT s + MT .eta. ( t ) x ~
( t - T 4 ) dt . ##EQU00006##
[0078] 7) Let .mu..sub..eta.E{.eta.(t)}, then
.mu. .eta. I = .DELTA. E { .eta. I [ n ] } = .intg. nT s nT s + MT
E { .eta. ( t ) } x ~ ( t ) dt = .mu. .eta. .intg. nT s nT s + MT x
~ ( t ) dt = 0 , .mu. .eta. Q = E { .eta. Q [ n ] } = .intg. nT s
nT s + MT E { .eta. ( t ) } x ~ ( t - T / 4 ) dt = .mu. .eta.
.intg. nT s nT s + MT x ~ ( t - T / 4 ) dt = 0 8 ) Let .sigma.
.eta. 2 = .DELTA. E { ( .eta. ( t ) - .mu. .eta. ) 2 } = E { (
.eta. ( t ) ) 2 } - .mu. .eta. 2 , then .sigma. .eta. I 2 = .DELTA.
E { ( .eta. I [ n ] ) 2 } = E { ( .intg. nT s nT s + MT .eta. ( t )
x ~ ( t ) dt ) ( .intg. nT s nT s + MT .eta. ( s ) x ~ ( s ) ds ) }
= E { .intg. nT s nT s + MT .intg. nT s nT s + MT .eta. ( t ) .eta.
( s ) x ~ ( t ) x ~ ( s ) dsdt } = .intg. nT s nT s + MT .intg. nT
s nT s + MT E { .eta. ( t ) .eta. ( s ) } x ~ ( t ) x ~ ( s ) dsdt
= .intg. nT s nT s + MT .intg. nT s nT s + MT .delta. ( s - t ) (
.sigma. .eta. 2 + .mu. .eta. 2 ) x ~ ( t ) x ~ ( s ) dsdt = (
.sigma. .eta. 2 + .mu. .eta. 2 ) .intg. nT s nT s + MT ( x ~ ( t )
) 2 dt = ( .sigma. .eta. 2 + .mu. .eta. 2 ) MT , .sigma. .eta. Q 2
= .DELTA. E { ( .eta. Q [ n ] ) 2 } = E { ( .intg. nT s nT s + MT
.eta. ( t ) x ~ ( t - T / 4 ) dt ) ( .intg. nT s nT s + MT .eta. (
s ) x ~ ( s - T / 4 ) ds ) } = E { .intg. nT s nT s + MT .intg. nT
s nT s + MT .eta. ( t ) .eta. ( s ) x ~ ( t - T / 4 ) x ~ ( s - T /
4 ) dsdt } = .intg. nT s nT s + MT .intg. nT s nT s + MT E { .eta.
( t ) .eta. ( s ) } x ~ ( t - T / 4 ) x ~ ( s - T / 4 ) dsdt =
.intg. nT s nT s + MT .intg. nT s nT s + MT .delta. ( s - t ) (
.sigma. .eta. 2 + .mu. .eta. 2 ) x ~ ( t - T / 4 ) x ~ ( s - T / 4
) dsdt = ( .sigma. .eta. 2 + .mu. .eta. 2 ) .intg. nT s nT s + MT (
x ~ ( t - T / 4 ) ) 2 dt = ( .sigma. .eta. 2 + .mu. .eta. 2 ) MT
##EQU00007##
[0079] According to the foregoing assumptions, the null hypothesis
(H.sub.0: EMPTY) in which I/Q discrete sampling has been applied is
expressed as Eqs. 13 and 14 below.
y I | 0 [ n ] = .intg. nT s nT s + MT y 0 ( t ) x ~ ( t ) dt =
.intg. nT s nT s + MT ( .varies. c A c x ( t - T c ) + .eta. ( t )
) x ~ ( t ) dt = .varies. c A c .intg. nT s nT s + MT x ( t - T c )
x ~ ( t ) dt + .intg. nT s nT s + MT .eta. ( t ) x ~ ( t ) dt = 2 P
.varies. c A c .intg. nT s nT s + MT ( p T / 2 ( t - T c ) *
.delta. T M ( t - nT s ) ) ( p T / 2 ( t ) * .delta. T M ( t - nT s
) ) dt - P .varies. c A c .intg. nT s nT s + MT p T / 2 ( t - T c )
* .delta. T M ( t - nT s ) dt + .eta. I [ n ] = 2 P .varies. c A c
.intg. nT s nT s + MT ( p T / 2 - T c ( t - T c ) * .delta. T M ( t
- nT s ) ) dt - P .varies. c A c ( T / 2 ) M + .eta. I [ n ] = 2 P
.varies. c A c ( T / 2 - T c ) M - P .varies. c A c ( T / 2 ) M +
.eta. I [ n ] = 2 P .varies. c A c ( T / 2 ) M - 2 P .varies. c A c
T c M - P .varies. c A c ( T / 2 ) M + .eta. I [ n ] = .varies. c A
c M P ( T / 2 - 2 T c ) + .eta. I [ n ] , ( Eq . 13 ) y Q | 0 [ n ]
= .intg. nT s nT s + MT y 0 ( t ) x ~ ( t - T / 4 ) dt = .intg. nT
s nT s + MT ( .varies. c A c x ( t - T c ) + .eta. ( t ) ) x ~ ( t
- T / 4 ) dt = .varies. c A c .intg. nT s nT s + MT x ( t - T c ) x
~ ( t - T / 4 ) dt + .intg. nT s nT s + MT .eta. ( t ) x ~ ( t - T
/ 4 ) dt = 2 P .varies. c A c .intg. nT s nT s + MT ( p T / 2 ( t -
T c ) * .delta. T M ( t - nT s ) ) ( p T / 2 ( t - T / 4 ) *
.delta. T M ( t - nT s ) ) dt - P .varies. c A c .intg. nT s nT s +
MT p T / 2 ( t - T c ) * .delta. T M ( t - nT s ) dt + .eta. Q [ n
] = 2 P .varies. c A c .intg. nT s nT s + MT ( p T / 2 - T / 4 - T
c ( t - T c ) * .delta. T M ( t - nT s ) ) dt - P .varies. c A c (
T / 2 ) M + .eta. Q [ n ] = 2 P .varies. c A c ( T / 2 - T / 4 - T
c ) M - P .varies. c A c ( T / 2 ) M + .eta. Q [ n ] = 2 P .varies.
c A c ( T / 2 ) M - 2 P .varies. c A c T / 4 - T c M - P .varies. c
A c ( T / 2 ) M + .eta. Q [ n ] = .varies. c A c M P ( T / 2 - 2 T
/ 4 - T c ) + .eta. Q [ n ] . ( Eq . 14 ) ##EQU00008##
[0080] According to the foregoing assumptions, the alternative
hypothesis (H.sub.1: NOT EMPTY) in which I/Q discrete sampling has
been applied is expressed as Eqs. 15 and 16 below.
y I | 1 [ n ] = .intg. nT s nT s + MT y 1 ( t ) x ~ ( t ) dt =
.intg. nT s nT s + MT ( .varies. b A b x ( t - T b ) + .varies. c (
A c - A b ) x ( t - T c ) + .eta. ( t ) ) x ~ ( t ) dt = .varies. b
A b .intg. nT s nT s + MT x ( t - T b ) x ~ ( t ) dt + .varies. c (
A c - A b ) .intg. nT s nT s + MT x ( t - T c ) x ~ ( t ) dt +
.intg. nT s nT s + MT .eta. ( t ) x ~ ( t ) dt ( Eq . 15 ) =
.varies. b A b M P ( T / 2 - 2 T b ) + .varies. c A c M P ( T / 2 -
2 T c ) - .varies. c A b M P ( T / 2 - 2 T c ) + .eta. I [ n ] =
.varies. c A b M P ( T / 2 - 2 T b ) - .varies. c A b M P ( T / 2 -
2 T c ) + y I | 0 [ n ] = A b M P ( .varies. b ( T / 2 - 2 T b ) -
.varies. c ( T / 2 - 2 T c ) ) + y I | 0 [ n ] , y Q | 1 [ n ] =
.intg. nT s nT s + MT y 1 ( t ) x ~ ( t - T / 4 ) dt = .intg. nT s
nT s + MT ( .varies. c A c x ( t - T b ) + .varies. c ( A c - A b )
x ( t - T c ) + .eta. ( t ) ) x ~ ( t - T / 4 ) dt = .varies. b A b
.intg. nT s nT s + MT x ( t - T b ) x ~ ( t - T / 4 ) dt + .varies.
c ( A c - A b ) .intg. nT s nT s + MT x ( t - T c ) x ~ ( t - T / 4
) dt + .intg. nT s nT s + MT .eta. ( t ) x ~ ( t - T / 4 ) dt ( Eq
. 16 ) = .varies. b A b M P ( T / 2 - 2 T / 4 - T b ) + .varies. c
A c M P ( T / 2 - 2 T / 4 - T c ) - .varies. c A b M P ( T / 2 - 2
T / 4 - T c ) + .eta. Q [ n ] = .varies. c A b M P ( T / 2 - 2 T /
4 - T b ) - .varies. c A b M P ( T / 2 - 2 T / 4 - T c ) + y Q | 0
[ n ] = A b M P ( .varies. b ( T / 2 - 2 T / 4 - T b ) - .varies. c
( T / 2 - 2 T / 4 - T c ) ) + y Q | 0 [ n ] ##EQU00009##
[0081] For readability, the following substitutions are used:
U.sub.bT/2-2T.sub.b, V.sub.bT/2-2|T/4-T.sub.b|;
(0.ltoreq.T.sub.b.ltoreq.T/2)
U.sub.cT/2-2T.sub.c, V.sub.cT/2-2|T/4-T.sub.c|;
(0.ltoreq.T.sub.c.ltoreq.T/2).
[0082] Thus, the received discrete I&Q samples for both H.sub.0
and H.sub.1are
H.sub.0: y.sub.1|0[n]=.varies..sub.cA.sub.CM {square root over
(P)}U.sub.c+.eta..sub.1[n]. (Eq. 17)
y.sub.Q|0[n]=.varies..sub.cA.sub.cA.sub.cM {square root over
(P)}U.sub.c+.eta..sub.Q[n].
H.sub.1: y.sub.1|1[n]=A.sub.bM {square root over
(P)}(.varies..sub.bU.sub.b31 .varies..sub.cU.sub.c)+y.sub.1|0[n].
(Eq. 18)
y.sub.Q|1[n]=A.sub.bM {square root over
(P)}(.varies..sub.bV.sub.b-.varies..sub.cV.sub.c)
+y.sub.Q|0[n].
[0083] For H.sub.1, both y.sub.1|1[n] and y.sub.Q|1[n] are in terms
of y.sub.1|0[n] and y.sub.Q|0[n], respectfully, along with
corresponding biasing terms for when cargo is present. In other
words, the above model describes the case where a physical item is
present using terms of when the container is empty along with the
adjustments to account for the physical item being present. One
interpretation is that the biasing terms provide a "delta offset"
to a "baseline" vector.
.DELTA..sub.l[n]A.sub.bM {square root over
(P)}(.varies..sub.bU.sub.b-.varies..sub.cU.sub.cU.sub.c). (Eq.
19)
.DELTA..sub.Q[n]A.sub.bM {square root over
(P)}(.varies..sub.bV.sub.b-.varies..sub.cV.sub.c). (Eq. 20)
[0084] In column-vector form, the discrete samples become
.DELTA.[n]=(.DELTA..sub.l[n].DELTA..sub.Q[n])'. (Eq. 21)
H.sub.0: y.sub.0[n]=(y.sub.l|0[n]y.sub.Q|0[n])'. (Eq. 22)
H.sub.1:
y.sub.1[n]=(y.sub.l|1[n]y.sub.Q|1[n])'=.DELTA.[n]+y.sub.0[n]; (Eq.
23)
[0085] Log-Likelihood Ratio (LLR) Test
[0086] A likelihood ratio test .lamda.(y) and the log-likelihood
ratio (LLR) test LLR(y) can be expressed as follows:
.lamda. ( y ) = .DELTA. f ( y | 1 ) f ( y | 0 ) 1 0 1 LLR ( y ) =
.DELTA. ln .lamda. ( y ) 1 0 0. ( Eq . 24 ) LLR ( y ) = .DELTA. ln
.lamda. ( y ) = ln f ( y | 1 ) - ln f ( y | 0 ) . ( Eq . 25 )
##EQU00010##
[0087] For the equi-variance multi-variable Gaussian noise model,
the following probability density distributions,
f ( y | 0 ) = .DELTA. 1 2 .pi. .SIGMA. exp ( - 1 2 ( y - .mu. 0 ) '
.SIGMA. - 1 ( y - .mu. 0 ) ) , ( Eq . 26 ) f ( y | 1 ) = .DELTA. 1
2 .pi. .SIGMA. exp ( - 1 2 ( y - .mu. 1 ) ' .SIGMA. - 1 ( y - .mu.
1 ) ) , ( Eq . 27 ) ##EQU00011##
represent an empty container and a loaded container, respectfully.
In Eq. 26, the parameter .mu..sub.0 represents an estimated mean of
the null hypothesis y.sub.0[n] signals, and in Eq. 27, the
parameter .mu..sub.1 represents an estimated mean of the
alternative hypothesis y.sub.1[n] signals. Note that y.sub.0[n]
represents received signals (such as signals received from the
light detector 404 corresponding to an empty container, and
y.sub.1[n] represents received signals corresponding to a non-empty
container. For an example channel model (model of a communications
medium through which light signals are transmitted and received)
and assuming zero-mean noise, the additive noise covariance matrix
can be expressed as
.SIGMA.E{.eta..eta.'}. (Eq. 28)
and the LLR function LLR(y) becomes
LLR(y)=1/2((y-.mu..sub.0)'.SIGMA..sup.-1(y-.mu..sub.0)-(y-.mu..sub.1)'.S-
IGMA..sup.-1(y-.mu..sub.1)) (Eq. 29)
[0088] If the additive noise is assumed to be additive white
Gaussian noise (AWGN), then
.SIGMA.=.sigma..sub..eta..sup.2I.sub.2, (Eq. 30)
and the LLR function LLR(y) simplifies further to
LLR ( y ) = 1 2 .sigma. .eta. 2 ( ( y - .mu. 0 ) ' ( y - .mu. 0 ) -
( y - .mu. 1 ) ' ( y - .mu. 1 ) ) = 1 2 .sigma. .eta. 2 ( ( y -
.mu. 0 ) ' ( y - .mu. 0 ) + ( .mu. 0 + .DELTA. - y ) ' ( y - .mu. 0
- .DELTA. ) ) = 1 2 .sigma. .eta. 2 ( 2 ( y ' - .mu. 0 ' ) .DELTA.
- .DELTA. ' .DELTA. ) , ( Eq . 31 ) ##EQU00012##
where
.mu..sub.1=.mu..sub.0+.DELTA., (Eq. 32)
[0089] Now, the log-likelihood ratio test can be written in terms
of a discrete signal
LLR ( y [ n ] ) = .DELTA. 1 2 .sigma. .eta. 2 ( 2 ( y ' [ n ] -
.mu. ^ 0 ' [ n ] ) .DELTA. ^ [ n ] - .DELTA. ^ ' [ n ] .DELTA. ^ [
n ] ) , ( Eq . 33 ) ##EQU00013##
where
{circumflex over (.mu.)}.sub.1[n]={circumflex over
(.mu.)}.sub.0[n]+{circumflex over (.DELTA.)}[n], (Eq. 34)
{circumflex over (.DELTA.)}[n]={circumflex over
(.mu.)}.sub.1[n]-{circumflex over (.mu.)}.sub.0[n] (Eq. 35)
[0090] In Eq. 34, {circumflex over (.mu.)}.sub.0[n] represents an
estimated baseline corresponding to an empty container.
[0091] FIG. 5A shows an example depicting values of the LLR
function LLR(y[n]) calculated as set forth above as data samples
y[n] are received, where n is the sample index represented by the
horizontal axis in FIG. 5A, and curve 502 represents the values of
LLR(y[n]) at each sample index.
[0092] As discussed further above, the output values of the
log-likelihood ratio test, LLR(y[n]), can be compared to thresholds
(including an upper threshold and a lower threshold) to determine
whether a container is empty or loaded. However, as seen in FIG.
5A, the LLR(y[n]) values vary widely as received signal samples are
continually received, and thus may not accurately determine the
load status of the container. As discussed further below, the
sequential probability ratio test (SPRT) can be used instead, which
aggregates (e.g., sums), multiple LLR(y[n]) values.
[0093] To set the upper and lower thresholds, consider
A < LLR ( y [ n ] ) < B ##EQU00014## A < 1 2 .sigma. .eta.
2 ( 2 ( y ' [ n ] - .mu. ^ 0 ' [ n ] ) .DELTA. ^ [ n ] - .DELTA. ^
' [ n ] .DELTA. ^ [ n ] ) < B ##EQU00014.2## .sigma. .eta. 2 A
< ( y ' [ n ] - .mu. ^ 0 ' [ n ] ) .DELTA. ^ [ n ] - .DELTA. ^ '
[ n ] .DELTA. ^ [ n ] / 2 < .sigma. .eta. 2 B ##EQU00014.3##
.sigma. .eta. 2 A < S ~ 0 < .sigma. .eta. 2 B ,
##EQU00014.4##
where {A, B} are real-numbered thresholds selected to control the
type II and type I error rates, respectively. Estimated thresholds
are calculated as follows, where .pi..sub.L represents the lower
threshold, and .pi..sub.U represents the upper threshold.
.pi..sub.L{circumflex over (.sigma.)}.sub..eta..sup.2[n]A,
.pi..sub.U{circumflex over (.sigma.)}.sub..eta..sup.2[n]B. (Eq.
36)
[0094] More generally, the upper and lower thresholds are computed
based on tuning parameters (such as A and B) selected to control
error rates.
[0095] In practice, obtaining a baseline estimate for {circumflex
over (.mu.)}.sub.1[n] may prove difficult because of load
dependency, so the process can start by assuming that the
observation y[n] is that of a loaded container, then
{circumflex over (.DELTA.)}[n]={circumflex over
(.mu.)}.sub.y[n]-{circumflex over (.mu.)}.sub.0[n], (Eq. 37)
can be used as the estimated "delta offset" where {circumflex over
(.mu.)}.sub.y[n] is the estimated sample mean of y[n].
H 0 : E { .DELTA. ^ [ n ] | 0 } = E { .mu. ^ y [ n ] - .mu. ^ 0 [ n
] } = E { .mu. ^ y [ n ] } - E { .mu. ^ 0 [ n ] } = .mu. 0 - .mu. 0
= 0 _ 2 .times. 1 , . ( Eq . 38 ) H 1 : E { .DELTA. ^ [ n ] | 1 } =
E { .mu. ^ y [ n ] - .mu. ^ 0 [ n ] } = E { .mu. ^ y [ n ] } - E {
.mu. ^ 0 [ n ] } = .mu. 1 - .mu. 0 = .DELTA. . ##EQU00015##
[0096] For a single received signal sample, {circumflex over
(.mu.)}.sub.n[n]=y[n], and
{circumflex over (.DELTA.)}[n]=y[n]-{circumflex over
(.mu.)}.sub.0[n]. (Eq. 39)
and the LLR expression simplifies to
LLR ( y ) = 1 2 .sigma. .eta. 2 ( .DELTA. I 2 + .DELTA. Q 2 ) . (
Eq . 40 ) ##EQU00016##
[0097] Going back to the case where the noise may be jointly
correlated, the log-likelihood ratio test is expressed as:
LLR(y)=1/2((y-.mu..sub.0)'.SIGMA..sup.-1(y-.mu..sub.0)'.SIGMA..sup.-1(y--
.mu..sub.1)), (Eq. 41)
for the LLR expression where the noise covariance matrix and its
inverse, respectively, are
.SIGMA. = ( .sigma. I 2 .rho. .sigma. I .sigma. Q .rho. .sigma. I
.sigma. Q .sigma. Q 2 ) ; .SIGMA. - 1 = ( .sigma. I - 2 - .rho.
.sigma. Q - 1 .sigma. I - 1 - .rho..sigma. I - 1 .sigma. Q - 1
.sigma. Q - 2 ) 1 1 - .rho. 2 . ( Eq . 42 ) ##EQU00017##
[0098] Using again, .mu..sub.1=.mu..sub.0+.DELTA. and
.DELTA.=.mu..sub.0-.mu..sub.1, the LLR expression simplifies to
LLR(y)=(y-.mu..sub.0)'.SIGMA..sup.-1.DELTA.-.DELTA.'.SIGMA..sup.-1.DELTA-
./2. (Eq. 43)
[0099] Again in practice, obtaining a baseline estimate for
{circumflex over (.mu.)}.sub.1[n] may prove difficult because of
load dependency, so the process can start by assuming that the
observation y[n] is that of a loaded container, then for a single
sample
{circumflex over (.DELTA.)}[n]=y[n]-{circumflex over
(.mu.)}.sub.0[n], (Eq. 44)
[0100] Thus, the LLR expression further simplifies to
LLR ( y ) = 1 2 ( 1 - .rho. 2 ) ( .DELTA. I 2 .sigma. I - 2 +
.DELTA. Q 2 .sigma. Q - 2 - 2 .DELTA. Q .DELTA. I .rho..sigma. I -
1 .sigma. Q - 1 ) . ( Eq . 45 ) ##EQU00018##
[0101] If .sigma..sub.Q=.sigma..sub.1=.sigma..sub..eta., then
LLR ( y ) = 1 2 ( 1 - .rho. 2 ) .sigma. .eta. 2 ( .DELTA. I 2 +
.DELTA. Q 2 - 2 .DELTA. Q .DELTA. I .rho. ) . ( Eq . 46 )
##EQU00019##
[0102] If .rho.=0, then
LLR ( y ) = 1 2 .sigma. .eta. 2 ( .DELTA. I 2 + .DELTA. Q 2 ) , (
Eq . 47 ) ##EQU00020##
which is the same as above when assuming AWGN.
[0103] Sequential Probability Ratio Test (SPRT):
[0104] The following describes the SPRT in detail, according to
some examples. As noted above, the SPRT sums or otherwise
aggregates a number of LLR output values to produce an SPRT output
value, and the SPRT output value (represented by curve 504 in FIG.
5B) can be compared to an upper threshold 506 and a lower threshold
508 as depicted in FIG. 5B. FIG. 5B depicts SPRT output values as a
function of received signal samples (sample indexes). Because the
SPRT output value is based on the sum of a number of LLR output
values, FIG. 5B shows the SPRT output value 504 as continually
increasing as additional received signal samples are received. When
the SPRT output value 504 exceeds the upper threshold 506, that
indicates that cargo has been detected, and the container can be
marked as loaded. In contrast, when the SPRT output value 504 drops
below the lower threshold 506, that indicates that the container is
empty.
[0105] Generally, in employing the SPRT, multiple values are
computed using a likelihood ratio test function (e.g., the LLR
function discussed above) applied on corresponding samples of
measurement data received from the at least one sensor. The
multiple values are aggregated to produce an aggregate value, and
then compared to at least one threshold to determine a load status
of a container.
[0106] Using the vector notation from earlier (see y.sub.0[n],
y.sub.1[n] and .DELTA.[n]), let
y[n](y.sub.1[n] y.sub.Q[n]), (Eq. 48)
be the received discrete coordinate samples, where
y.sub.l[n].intg..sub.nT.sub.s.sup.nT.sup.S.sup.+MTy(t){tilde over
(x)}(t)dt,
y.sub.Q[n].intg..sub.nT.sub.s.sup.nT.sup.s.sup.+MTy(t){tilde over
(x)}(t-T/4) dt. (Eq. 49)
[0107] Further expanding on this notation,
y[0: n](y.sub.1[0: n] y.sub.Q[0: n]), (Eq. 50)
y.sub.l[0: n](y.sub.l[0] y.sub.l[1] . . . y.sub.l[n])', (Eq.
51)
y.sub.Q[0: n](y.sub.Q[0]y.sub.Q[1] . . . y.sub.Q[n])', (Eq. 52)
with each including the latest (n+1) samples.
.lamda. ( y [ 0 : n ] ) = .DELTA. f ( y [ 0 : n ] | 1 ) f ( y [ 0 :
n ] | 0 ) = k = 0 n f ( y [ k ] | 1 ) k = 0 n f ( y [ k ] | 0 ) , (
Eq . 53 ) LLR ( y [ 0 : n ] ) = .DELTA. ln .lamda. ( y [ 0 : n ] )
= k = 0 n f ( y [ k ] | 1 ) - k = 0 n f ( y [ k ] | 0 ) = k = 0 n
LLR ( y [ k ] ) , ( Eq . 54 ) S n = .DELTA. k = 0 n LLR ( y [ k ] )
= k = 0 n 1 2 .sigma. .eta. 2 ( 2 ( y ' [ k ] - .mu. ^ 0 ' [ k ] )
.DELTA. ^ [ k ] - .DELTA. ^ ' [ k ] .DELTA. ^ [ k ] ) . ( Eq . 55 )
##EQU00021##
[0108] In Eq. 55, S.sub.n represents the SPRT output value derived
from summing n+1 LLR output values.
[0109] In order to set the upper and lower thresholds (represented
as 506 and 508 in FIG. 5B), consider
A < S n < B ##EQU00022## A < 1 2 .sigma. .eta. 2 k = 0 n (
2 ( y ' [ k ] - .mu. ^ 0 ' [ k ] ) .DELTA. ^ [ k ] - .DELTA. ^ ' [
k ] .DELTA. ^ [ k ] ) < B ##EQU00022.2## .sigma. .eta. 2 A <
k = 0 n ( y ' [ k ] - .mu. ^ 0 ' [ k ] ) .DELTA. ^ [ k ] - k = 0 n
.DELTA. ^ ' [ k ] .DELTA. ^ [ k ] / 2 < .sigma. .eta. 2 B
##EQU00022.3## .sigma. .eta. 2 A < S ~ n < .sigma. .eta. 2 B
, ##EQU00022.4##
where {A, B} are real-numbered thresholds selected to control the
type II and type I error rates, respectively, and
S ~ n = .DELTA. k = 0 n ( ( y ' [ k ] - .mu. ^ 0 ' [ k ] ) .DELTA.
^ [ k ] - .DELTA. ^ ' [ k ] .DELTA. ^ [ k ] 2 ) , ( Eq . 56 )
##EQU00023##
[0110] In practice, implementations can use the sampled statistic
{circumflex over (.sigma.)}.sub..eta..sup.2[n] to estimate
.sigma..sub..eta..sup.2 which inherently has a variance, so to
mitigate early termination due to small sample size variation, a
process can "gear-step" the upper and lower thresholds using any
number of schemes, e.g.,
.pi. _ L < S ~ n < .pi. _ U - .infin. < S ~ n < +
.infin. ; n < N .infin. ##EQU00024## .sigma. ^ .eta. 2 [ n ] ( A
- Ce ( 1 - n ) ) < S ~ n < .sigma. ^ .eta. 2 [ n ] ( B + De (
1 - n ) ) ; n .gtoreq. N .infin. ##EQU00024.2##
where {A,B,C,D} are real-numbered constants selected to control the
type I and type II error rates and N.sub..infin. is the number of
samples to wait allowing for settlement of the sample statistic
variances (i.e., minimum number of observed samples before allowing
detection to make a decision). Thus, the lower and upper thresholds
(.pi..sub.L, .pi..sub.U) are computed according to Eq. 57:
.pi. _ L = .DELTA. { .sigma. ^ .eta. 2 [ n ] ( A - Ce ( 1 - n ) ) ;
n .gtoreq. N .infin. - .infin. ; n < N .infin. and .pi. _ U =
.DELTA. { .sigma. ^ .eta. 2 [ n ] ( A - Ce ( 1 - n ) ) ; n .gtoreq.
N .infin. + .infin. ; n < N .infin. , ( Eq . 57 )
##EQU00025##
[0111] Another example of "gear-stepping" the upper and lower
thresholds can be according to
.pi. _ L < S ~ n < .pi. _ U - .infin. < S ~ n < +
.infin. ; n < N .infin. ##EQU00026## .sigma. ^ .eta. 2 [ n ] ( A
- C n ) < S ~ n < .sigma. ^ .eta. 2 [ n ] ( B + D n ) ; n
.gtoreq. N .infin. ##EQU00026.2##
where
.pi. _ L = .DELTA. { .sigma. ^ .eta. 2 [ n ] ( A - C n ) ; n
.gtoreq. N .infin. - .infin. ; n < N .infin. and .pi. _ U =
.DELTA. { .sigma. ^ .eta. 2 [ n ] ( B + D n ) ; n .gtoreq. N
.infin. + .infin. ; n < N .infin. , ( Eq . 58 ) ##EQU00027##
are the corresponding lower and upper thresholds, respectfully.
[0112] Generalized-Likelihood-Ratio-Test (GLRT):
[0113] Use of the LLR and SPRT tests discussed above assume that
certain characteristics of the container are known, such as the
color of the inner wall of the container. In some cases, such
characteristics of the container may not be known. As a result, the
parameter {circumflex over (.mu.)}.sub.0[n] that represents an
estimated mean of the null hypothesis y.sub.0[n] signals cannot be
derived, which would mean that the LLR and SPRT tests cannot be
used.
[0114] In such situations, the GLRT can be used instead, where the
GLRT can use a different value to represent {circumflex over
(.mu.)}.sup.0[n], as explained below.
[0115] Let an observed received signal at time n be y[n] as
follows:
y [ n ] = ( y I [ n ] y Q [ n ] ) ; d [ n ] = r d [ n ] ( cos
.theta. d sin .theta. d ) , ( Eq . 59 ) ##EQU00028##
where
.theta. d = tan - 1 V c U c = tan - 1 ( T / 2 - 2 T / 4 - T c T / 2
- 2 T c ) ; ( for square - wave basis functions ) ( Eq . 60 )
.theta. d = 4 .pi. L c Tc m / s ; ( for sinusoidal basis functions
) ( Eq . 61 ) ##EQU00029##
[0116] and d[n] is the point along the subspace defined by the
phase .theta..sub.d in the I/Q plane (an I/Q plane is shown in FIG.
6) that is r.sub.d[n] from the origin where the round-trip ToF
defines .theta..sub.d in radians for the given carrier period T and
the speed of light c.sub.m/s. In Eq. 60, T.sub.c is the round-trip
time shift attributed to the container's reflection, and U.sub.c
and V.sub.c are defined right above Eq. 16 further above. In Eq.
61, L.sub.c represents the length of the container.
[0117] In FIG. 6, the horizontal axis of the I/Q plane is the
in-phase (I) axis, while the vertical axis is the quadrature-phase
(Q) axis. Using the Cartesian distance for the norm, a vector
y[n]-d[n] (represented by line 600 in FIG. 6) is the vector
connecting the points y[n] and d[n]. A point 602 on the line 600
represents d[n]. The angle .theta..sub.d is between the I axis and
the line 600.
[0118] FIG. 6 further shows points 606 (blank circles) that are
observed received signals y[n] corresponding to signals of a
container including at least one physical item. A darkened circle
608 is a mean (or other aggregate) of the observed received signals
y[n] corresponding to signals of a container including at least one
physical item. This mean is referred to as a "mean observation" in
the ensuing discussion. The dashed oval 610 around the line 600
represents the decision region of the GLRT. The origin endpoint 612
of the line 600 represents a container with a black wall, where no
light is reflected. The opposite endpoint 614 of the line 600
represents a container with a white wall, where all light is
reflected. Other points along the line 600 represent other colors
of the wall of the container.
[0119] The GLRT seeks to find the closest point along the line 600
to the mean observation (608). In FIG. 6, such closest point is the
point 602 representing d[n]. This extrema point (602) is unique, so
the GLRT can take the derivative of the vector y[n]-d[n] with
respect to r.sub.d[n],
y [ n ] - d [ n ] 2 2 = y I [ n ] - r d [ n ] cos .theta. d y Q [ n
] - r d [ n ] sin .theta. d 2 2 = ( y I [ n ] - r d [ n ] cos
.theta. d ) 2 + ( y Q [ n ] - r d [ n ] sin .theta. d ) 2 ( Eq . 62
) .differential. .differential. r d [ n ] y [ n ] - d [ n ] 2 2 =
.differential. .differential. r d [ n ] ( y I [ n ] - r d [ n ] cos
.theta. d ) 2 + .differential. .differential. r d [ n ] ( y Q [ n ]
- r d [ n ] sin .theta. d ) 2 .differential. .differential. r d [ n
] ( y I [ n ] - r d [ n ] cos .theta. d ) 2 = - 2 y I [ n ] cos
.theta. d + 2 r d [ n ] ( cos .theta. d ) 2 .differential.
.differential. r d [ n ] ( y Q [ n ] - r d [ n ] sin .theta. d ) 2
= - 2 y Q [ n ] sin .theta. d + 2 r d [ n ] ( sin .theta. d ) 2 . (
Eq . 63 ) ##EQU00030##
The derivative is set to zero, and the GLRT can solve for
r.sub.d[n] as follows:
.differential. .differential. r d [ n ] y [ n ] - d [ n ] 2 2 = - 2
y I [ n ] cos .theta. d + 2 r d [ n ] ( cos .theta. d ) 2 - 2 y Q [
n ] sin .theta. d + 2 r d [ n ] ( sin .theta. d ) 2 = 0 , ( Eq . 64
) r d [ n ] ( ( cos .theta. d ) 2 + ( sin .theta. d ) 2 ) = y I [ n
] cos .theta. d + y Q [ n ] sin .theta. d , ( Eq . 65 ) r d [ n ] =
y I [ n ] cos .theta. d + y Q [ n ] sin .theta. d . ( Eq . 66 )
##EQU00031##
[0120] Note that r.sub.d[n] may be negative depending on the
combination of observation y[n] and the container's length L.sub.c.
Because of the physical interpretation, d[n] can be restricted to
be the origin when r.sub.d[n] is negative, i.e.,
d ^ [ n ] = r ^ d [ n ] ( cos .theta. d sin .theta. d ) where r ^ d
[ n ] = .DELTA. { r d [ n ] ; 0 .ltoreq. r d [ n ] 0 ; 0 > r d [
n ] . ( Eq . 67 ) ##EQU00032##
[0121] In accordance with some implementations of the present
disclosure, the parameter d[n] calculated above can be used as the
baseline vector {circumflex over (.mu.)}.sub.0[n]. Given this
estimate of {circumflex over (.mu.)}.sub.0[n], either the LLR test
or the SPRT can be then used, as discussed above.
[0122] More generally, the GLRT involves defining a subspace (e.g.
subspace 610 in FIG. 6) based on a phase (e.g., .theta..sub.d),
identifying a point (e.g., d[n]) in the subspace that is closest to
points representing measurement data received from at least one
sensor, using the identified point as an estimated baseline
corresponding to an empty container, and computing a value using
the estimated baseline, where the value is computed by using an LLR
test or the SPRT.
[0123] In some examples as discussed above, the load status
detector 208 and the LRT function 209 (which can perform any of the
LLR, SPRT, or GLRT tests, for example) can be implemented as
machine-readable instructions executable on one or more
processors.
[0124] The machine-readable instructions can be stored in a
non-transitory computer-readable or machine-readable storage
medium. The storage medium can include one or multiple different
forms of memory including semiconductor memory devices such as
dynamic or static random access memories (DRAMs or SRAMs), erasable
and programmable read-only memories (EPROMs), electrically erasable
and programmable read-only memories (EEPROMs) and flash memories;
magnetic disks such as fixed, floppy and removable disks; other
magnetic media including tape; optical media such as compact disks
(CDs) or digital video disks (DVDs); or other types of storage
devices. Note that the instructions discussed above can be provided
on one computer-readable or machine-readable storage medium, or
alternatively, can be provided on multiple computer-readable or
machine-readable storage media distributed in a large system having
possibly plural nodes. Such computer-readable or machine-readable
storage medium or media is (are) considered to be part of an
article (or article of manufacture). An article or article of
manufacture can refer to any manufactured single component or
multiple components. The storage medium or media can be located
either in the machine running the machine-readable instructions, or
located at a remote site from which machine-readable instructions
can be downloaded over a network for execution.
[0125] In the foregoing description, numerous details are set forth
to provide an understanding of the subject disclosed herein.
However, implementations may be practiced without some of these
details. Other implementations may include modifications and
variations from the details discussed above. It is intended that
the appended claims cover such modifications and variations.
* * * * *